EP0654771A1 - Method for preventing false alarms in a fire detecting system and device for performing this method - Google Patents

Abstract

The fire-detecting system contains detectors with sensors (1) for monitoring fire characteristics. The sensors (1) generate corresponding sensor signals (S) which are fed to an analytical stage (evaluation stage) (3). The probability (W) of a future false alarm is estimated in this analytical stage and an indication signal is output if the probability has a specific magnitude. <IMAGE>

Description

The present invention relates to a method for avoiding false alarms in a fire alarm system, with a plurality of detectors connected to a control center, which have at least one sensor for monitoring fire parameters and for emitting corresponding sensor signals, from which hazard signals are derived in a signal processing process.

False alarms, which are among the most common faults in fire protection systems, occur, among other things, because the sensors are "mistaken" in that they cannot distinguish between a fire parameter and a parameter that only pretend a fire. The main reason for this deception is that both sizes are physically the same but of different origin, that is to say the physical size "smoke" can be caused, for example, once by a fire, once by a cigar smoker and once by welding work in the respective room. If the detector in question responds to the smoke parameter, it will do so in each of the three cases, and it will not be possible for the cigar smoker to trigger false alarms or the welding work to be carried out by increasing the reliability of the sensor or individual components of to prevent this. However, since known systems are aimed almost exclusively at such an improvement in reliability, they cannot generally reduce the number of false alarms of the type described.

The invention is now intended to provide a method by which false alarms are largely avoided or at least noticeably reduced through its use.

• a. Analyse der Sensorsignale während eines bestimmten ersten Intervalls,
• b. Berechnung der Wahrscheinlichkeit eines Fehlalarms in einem folgenden zweiten Intervall; und
• c. Abgabe eines Hinweises, wenn die Wahrscheinlichkeit einen bestimmten Wert überschreitet.

According to the invention, this object is achieved in that the signal processing process contains the following steps:

• a. Analysis of the sensor signals during a certain first interval,
• b. Calculating the probability of a false alarm in a subsequent second interval; and
• c. Providing information when the probability exceeds a certain value.

In order to avoid false alarms, the method according to the invention therefore takes a completely different route than before: attempts are not made to reduce the false alarms by increasing the reliability of the system or its components, but rather the system is designed in such a way that false alarms can be predicted. If the probability of a future false alarm reaches or exceeds a certain value, the user receives a message or warning and can react accordingly.

A major difficulty with such a method or system lies in the relationship between the time it takes to decide whether or not to issue a warning and the reliability of that decision. On the one hand, the decision must be made within the shortest possible time, since a false alarm usually takes place shortly after a change in the ambient conditions. On the other hand, the statistical relevance of the data collected during this short period of time is not great and cannot be.

This problem of estimating the probability of a false alarm on the basis of only a little information is solved according to a preferred development of the method according to the invention in that the length of the second time segment is of the same order of magnitude as that of the first, that each time segment is divided into subintervals and for each subinterval the mean value of the signal maxima is determined, and that the distribution function of the probability of a false alarm is derived from this mean value.

One of the main uses of the method according to the invention is in the so-called application error determination, by means of which the user is to be made aware of possible application errors. According to another preferred development of the method according to the invention, this function can be fulfilled in that instead of calculating the probability according to step b. a threshold value is determined, that the sensor signals are compared with this threshold value and the exceedances of the threshold value are registered, and that an application error message is given if a certain number of exceedances is exceeded.

The invention further relates to a fire detection system for carrying out the above-mentioned method, with a control center, with detectors connected to it, which have sensors for fire parameters and emit corresponding sensor signals, and with means for processing them.

The fire alarm system according to the invention is characterized in that said means for processing the sensor signals have means for their registration during the first interval, means for comparing the sensor signals with a threshold value and means for registering the exceeding of the threshold value by the sensor signals.

• Fig. 1 ein Blockschema der Signalverarbeitung; und
• Fig. 2 ein Diagramm zur Erläuterung einer speziellen Funktion, der sogenannten Applikationsfehlerermittlung.

The invention will now be explained in more detail with reference to exemplary embodiments and the drawings; shows:

• 1 shows a block diagram of the signal processing; and
• Fig. 2 is a diagram for explaining a special function, the so-called application error determination.

In Fig. 1, reference numeral 1 designates the or a sensor of a fire alarm, at the output of which a sensor signal S is available. The reference numeral 2 denotes a block 2 in which the quantization of the sensor signals S takes place, that is to say the continuous sensor signal is sampled. Reference number 3 denotes a stage for signal analysis, at the output of which is the probability of one False alarm indicating signal W is available. The signal analysis usually does not take place in the detector but in the control center to which the detectors with sensors 1 are connected. It does not matter whether the control center receives the sensor signal S in quantized form or not; in the latter case, the quantization would take place in the control center, which is indicated in the figure by the dashed line connecting sensor 1 with analysis stage 3.

In analysis stage 3, an interval is first defined over which the sensor signal is to be analyzed. The length of this interval can range from minutes, days, weeks or even months. Preferably, not only an interval is specified, but a series of intervals with different lengths. The latter is done by dividing the intervals into subintervals, and so on, so that an interval grid is usually available, in whose individual differently rastered subintervals the sensor signal is analyzed.

A second interval of preferably the same length as the first or an interval grid with lengths corresponding to the first interval grid is then defined and the result of the analysis of the sensor signal in the individual subintervals of the first interval is transferred to the corresponding subintervals of the second interval. In such a way that one examines whether an indication can be derived from the behavior or course of the signal in a first interval, that a false alarm could be triggered in the corresponding second interval, and how high the probability is.

A main prerequisite for being able to infer the behavior in a second interval from the behavior of the sensor signal S is the presence of a steady state. It is therefore assumed that steady-state conditions prevailed during the observation and registration of the signal, and that this will continue to be the case in the future during the second interval.

The definition of intervals of different lengths is recommended because the weighting of a signal with regard to its relevance for a possible one Alarm is very dependent on the time reference. If, for example, 20 events occur on a single day, i.e. exceeding any threshold value, then these are 20 independent events in relation to an interval the length of a day. In relation to a half-year or annual interval, on the other hand, it is a cluster of events that can in no way be regarded as independent of one another.

So that an event is not counted more than once, only the event with the greatest amplitude per subinterval is taken into account when considering the intervals composed of several subintervals in analysis stage 3. Although this means that all events with amplitudes below the maximum are not taken into account in a given subinterval, it is not critical because these events are detected in shorter intervals and subintervals. An average value representative of the respective interval is then formed from the maximum values of the individual subintervals. The probability of a false alarm is ultimately derived from this mean.

berechnet, dann ist die Wahrscheinlichkeit P für einen Fehlalarm während eines Subintervalls m und für einen gegebenen Schwellwert L gegeben durch

$\text{P (T/m, L) =∫λ exp (-λx)= exp (-λL) (Integral von L bis unendlich).}$

If one starts from the assumption that the distribution function of this probability is an exponential function, and if one divides an interval of length T into subintervals and from the mean of the signal maxima in the subintervals the parameter λ of the standardized distribution function $\text{f (λ, x) = λ exp (- λx)}$

is calculated, then the probability P for a false alarm during a subinterval m and for a given threshold value L is given by

$\text{P (T / m, L) = ∫λ exp (-λx) = exp (-λL) (integral from L to infinity).}$

und während des gesamten Intervalls:

$\overline{\text{P}}{\text{(T, L) = (1 - e}}^{\text{-λL}}{\text{)}}^{\text{m}}$

In der Praxis legt der Benutzer fest, bis zu welchem Ausmass das System Fehlalarme verhindern soll. Wenn beispielsweise 9 von 10 Fehlalarmen verhindert werden sollen, dann setzt man P gleich 0.9. Dieser Wert und die Anzahl m der Subintervalle definiert die Bedingung für die Abgabe einer Warnung durch das System:
Warnung, wenn:

${\text{λL ≧ - In[1 - P}}^{\text{1/m}}\text{(T,L)]}$

Für P = 0.9 und 10 Subintervalle berechnet sich das Verhältnis Schwellwert L zu Mittelwert 1/λ zu:

${\text{λL = - In(1 - 0.9}}^{\text{1/10}}\text{) = 4.55}$

Dieses Ergebnis besagt, dass der Mittelwert der in einem gegebenen Intervall gesammelten Daten 22% des Alarm-Schwellwerts nicht übersteigen soll, wenn das System innerhalb des nächsten Intervalls derselben Länge einen Fehlalarm mit der Wahrscheinlichkeit 0.9 vermeiden soll.The probability of avoiding a false alarm during a subinterval is:

$\overline{\text{P}}{\text{(T / m, L) = 1 - e}}^{\text{-λL}}$

and throughout the interval:

$\overline{\text{P}}{\text{(T, L) = (1 - e}}^{\text{-λL}}{\text{)}}^{\text{m}}$

In practice, the user determines to what extent the system should prevent false alarms. For example, if 9 out of 10 false alarms are to be prevented, P is set to 0.9. This value and the number m of subintervals define the condition for the system to issue a warning:
Warning if:

${\text{λL ≧ - In [1 - <figure-callout id="1" label="P" filenames="imgf0001.png" state="{{state}}">P</figure-callout>}}^{\text{1 / m}}\text{(T, L)]}$

For P = 0.9 and 10 subintervals, the ratio threshold L to mean 1 / λ is calculated:

${\text{λL = - In (1 - 0.9}}^{\text{1/10}}\text{) = 4.55}$

This result means that the average of the data collected in a given interval should not exceed 22% of the alarm threshold if the system is to avoid a false alarm with probability 0.9 within the next interval of the same length.

In practical implementation, the bandwidth of the intervals will be chosen such that the shortest is defined by the shortest reaction time of a user, typically 10 minutes, and the longest by the maximum expected duration of the steady state, for example 6 months. If you double the interval length based on the shortest interval, you get 15 intervals from 10 minutes to 6 months. The mean values for each interval are obtained by filtering the maxima of the subintervals with a digital low-pass filter. This mean value is saved together with the provisional maximum for each interval.

The algorithm for the warning is very simple: the system calculates the mean values and checks whether they exceed a given threshold value corresponding to the probability of avoiding a false alarm P. This can be different for each interval. If, as stated above, 9 out of 10 false alarms are to be prevented, the system will, as soon as it detects that the mean value becomes one within an interval of, for example, one hour Has exceeded 22% of the threshold, give a warning and request intervention within the next hour. If the interval is 1 month, the message would look different because the intervention would not be so urgent.

2 shows an exemplary embodiment for a very simple function of the method according to the invention. This function is a so-called application error determination or message, which is intended to draw the user's attention to any application errors. The basic idea is that it is automatically determined whether and how often a detector exceeds a certain danger level that does not yet trigger an alarm within a certain interval. Because then there is a risk that a false alarm will be triggered at some point.

The upper half of FIG. 2 shows the diagram of a sensor signal S plotted over time t, a threshold value G1 for the aforementioned low danger level being plotted on the ordinate. A detector counts each time the threshold value G1 is exceeded and delivers a corresponding pulse In to a counter 4. The counter 4 counts the pulses In over the selected time interval T of, for example, 24 hours and reports the counter reading, which is 5 in the example shown, to a comparator 5 at the end of the time interval. This compares the counter reading received with a set value and, when this value is exceeded, issues an advisory message of the "unsuitable use" type or the like.

The exemplary embodiment shown can be expanded further, for example by quantizing the signal S and thus determining how long it took the signal S to exceed the threshold value G1. Of course, the other, higher danger levels for the application error determination can also be taken into account by taking into account that these danger levels are exceeded for the notification message.

Claims (10)

Method for avoiding false alarms in a fire detection system, with a plurality of detectors connected to a control center, which have at least one sensor for monitoring fire parameters and for emitting corresponding sensor signals, from which hazard signals are derived in a signal processing process, characterized in that the signal processing process follows Steps includes: a. Analysis of the sensor signal (S) during a specific first interval, b. Calculating the probability (W) of a false alarm in a subsequent second interval; and c. Providing information when the probability exceeds a certain value. A method according to claim 1, characterized in that the second interval is chosen to be approximately the same length as the first. A method according to claim 2, characterized in that the sensor signal (S) is analyzed over several intervals of different lengths. Method according to one of claims 1 to 3, characterized in that each interval is divided into a number of subintervals of equal length. A method according to claim 4, characterized in that the maximum value of the sensor signal (S) is determined in each subinterval and that an average value for the relevant interval is calculated from the maximum values of all subintervals. A method according to claim 5, characterized in that, based on the extent to which false alarms are to be prevented, a threshold value is defined, the indication of which is given when the average value is exceeded. A method according to claim 1, characterized in that instead of calculating the probability according to step b. a threshold value (G1) is determined, that the sensor signals (S) are compared with this threshold value, and an application error message is given if a certain number of violations is exceeded. Fire alarm system for carrying out the method according to claim 1 or 7, with a control center, with detectors connected to it, which have sensors for fire parameters and emit corresponding sensor signals, and with means for their processing, characterized in that the said means for processing the sensor signals ( S) have means for their registration during the first interval, means for comparing the sensor signals with a threshold value (G1) and means (4) for registering the exceeding of the threshold value by the sensor signals. Fire alarm system according to Claim 8, characterized in that the means for processing the sensor signals (S) contain means for dividing the first interval into subintervals of equal length, means for determining the maxima of the sensor signal in the subintervals and means for forming an interval mean value from the maxima. Fire alarm system according to claim 9, characterized in that the means for forming the interval mean are formed by a digital low-pass filter.

Images